Metastasis is the major cause of death in cancer patients, but the molecular basis of metastatic cancer is poorly understood. This likely is due to the genetic complexity of the metastatic phenotype, which is not easily studied using traditional methods. We have used oligonucleotide microarrays to define gene expression differences among human cancers, with an eye towards identifying molecular markers that are biologically and clinically informative. Recently, we compared primary and metastatic tumors from a variety of different sites, reasoning that this approach would allow us to identify genes that are differentially expressed in the metastatic state. Indeed, this strategy did allow us to define a "metastasis" gene expression signature that is commonly expressed in different types of metastatic cancer. Remarkably, this molecular signature is also expressed in some primary tumors but not others, and the presence of this signature in a primary tumor portends worse clinical prognosis in multiple solid tumor types. These findings suggest that the metastatic potential of a primary tumor is pre-configured, in contrast to the long-held view that rare metastatic cells can arise stochastically in any primary tumor. Component genes of this metastasis signature might also play functional roles in causing metastasis that transcend tissue-specific differences. These findings warrant further study, to develop the best possible clinical cancer diagnostics and to gain more biological insight into the molecular basis of metastasis. We propose a multifaceted and interdisciplinary set of studies. First, our initial findings resulted from the analysis of a relatively small number of metastatic tumor specimens. Analysis of a larger set of metastatic tumors should allow us to gain even greater statistical confidence in our observation. We therefore will obtain high-quality microarray-based gene expression data from a larger, specially curated set of metastatic human tumors. Next, we will perform a rigorous analysis of the resulting gene expression dataset, to identify an information-rich metastasis gene set in a statistically principled manner. Finally, we will perform histopathologic and functional experiments to define the biological role that these genes might have in causing cancer cell metastasis. These studies should enable us to (1) define novel molecular diagnostics that are predictive of metastatic disease in solid tumors and (2) identify mechanistically important metastasis genes.